[Wake/VAD/STT]：完成本地唤醒、人声端点检测与转写入口，包含小杰小杰唤醒、VAD 录音切分、Metadata STT 和 sherpa-onnx 错误边界

src/owner_voice_pet/stt.py

@@ -0,0 +1,93 @@
+from __future__ import annotations
+
+import re
+from pathlib import Path
+
+from .models import AudioSegment, ErrorCode, ProviderError, Transcript
+
+_MEANINGFUL_TEXT = re.compile(r"[\w\u4e00-\u9fff]", re.UNICODE)
+
+
+def is_valid_transcript_text(text: str) -> bool:
+    return bool(_MEANINGFUL_TEXT.search(text.strip()))
+
+
+class MetadataSttProvider:
+    def __init__(self, language: str = "zh") -> None:
+        self.language = language
+        self.loaded = False
+
+    def load(self) -> None:
+        self.loaded = True
+
+    def transcribe(self, segment: AudioSegment) -> Transcript:
+        if not self.loaded:
+            raise ProviderError(
+                ErrorCode.STT_TRANSCRIBE_FAILED,
+                "STT provider is not loaded",
+                False,
+                "metadata-stt",
+                "stt",
+            )
+        text = str(segment.metadata.get("transcript", "")).strip()
+        if not is_valid_transcript_text(text):
+            raise ProviderError(
+                ErrorCode.STT_EMPTY_TRANSCRIPT,
+                "STT produced no meaningful text",
+                True,
+                "metadata-stt",
+                "stt",
+            )
+        return Transcript(
+            text=text,
+            language=str(segment.metadata.get("language", self.language)),
+            confidence=float(segment.metadata.get("stt_confidence", 1.0)),
+            duration_ms=segment.duration_ms,
+            provider="metadata-stt",
+            raw_metadata=dict(segment.metadata),
+        )
+
+
+class SherpaOnnxSttProvider:
+    def __init__(self, model_path: str, language: str = "zh") -> None:
+        self.model_path = Path(model_path)
+        self.language = language
+        self.loaded = False
+
+    def load(self) -> None:
+        if not self.model_path.exists():
+            raise ProviderError(
+                ErrorCode.STT_MODEL_MISSING,
+                f"sherpa-onnx STT model path does not exist: {self.model_path}",
+                False,
+                "sherpa-onnx-stt",
+                "stt",
+            )
+        try:
+            import sherpa_onnx  # type: ignore[import-not-found]  # noqa: F401
+        except Exception as exc:
+            raise ProviderError(
+                ErrorCode.STT_TRANSCRIBE_FAILED,
+                f"sherpa_onnx is not available: {exc}",
+                False,
+                "sherpa-onnx-stt",
+                "stt",
+            ) from exc
+        self.loaded = True
+
+    def transcribe(self, segment: AudioSegment) -> Transcript:
+        if not self.loaded:
+            raise ProviderError(
+                ErrorCode.STT_TRANSCRIBE_FAILED,
+                "sherpa-onnx STT provider is not loaded",
+                False,
+                "sherpa-onnx-stt",
+                "stt",
+            )
+        raise ProviderError(
+            ErrorCode.STT_TRANSCRIBE_FAILED,
+            "sherpa-onnx runtime transcription adapter requires a concrete model profile",
+            False,
+            "sherpa-onnx-stt",
+            "stt",
+        )